[Rocks-Discuss] Journal commit i/o error & Input/Output error

Andreas Yiotis yiotis at ipta.demokritos.gr
Tue Jan 15 06:38:24 PST 2008


I finally had a chance to take another look at this problem and I am now
pretty convinced that you are right.

It seems that this hard disk WD5000AAKS is not fully supported by the
linux kernel and it enters some kind

of power-save mode that results in a journal commit I/O error.
Unfortunately, WD does not

provide any firmware update for this drive so I decided to replace the
drives.

Thanx for the tip Larry.

 

 

________________________________

From: Larry Baker [mailto:baker at usgs.gov] 
Sent: Wednesday, December 05, 2007 8:48 PM
To: Andreas Yiotis
Cc: Robert E. Parrott; bruno at rocksclusters.org;
npaci-rocks-discussion at sdsc.edu
Subject: Re: [Rocks-Discuss] Journal commit i/o error & Input/Output
error

 

I had a similar, seemingly random disk problem for about a year.  Not
the boot disk, though.  I was using two Western Digital disks as a Linux
RAID 1 mirrored pair.  It turned out the WD firmware put the drives into
a low-power sleep mode that Linux was not prepared to deal with.  They
had an updated firmware I downloaded that removed the feature.  However,
by then I had already taken the drives out of service and replaced them
with Seagate drives, which have been fine.  You might have a similar
problem with your drives.  Look at the manufacturer's web site for
either replacement firmware or a utility to configure the drive
parameters to disable this feature.  Otherwise, try another brand of
disk.

 

Larry Baker

US Geological Survey

650-329-5608

baker at usgs.gov

 

On Dec 5, 2007, at 2:10 AM, Andreas Yiotis wrote:





Unfortunately, the problem "journal commit I/O error" has re-appeared

overnight. It can't be the HD because I have changed it yesterday.

I am pretty sure now that the size (500GB) or type of the HD is the

problem.

I tried with a smaller 160GB IDE disk yesterday (reinstalled rocks) and

the

System works fine so far.

Any ideas about how to make the large Disk working with rocks?

 

A

 

 

 

-----Original Message-----

From: Robert E. Parrott [mailto:parrott at seas.harvard.edu] 

Sent: Tuesday, December 04, 2007 6:39 PM

To: Andreas Yiotis

Cc: bruno at rocksclusters.org; npaci-rocks-discussion at sdsc.edu

Subject: Re: [Rocks-Discuss] Journal commit i/o error & Input/Output

error

 

I've seen this problem before on a VMWare virtual machine, and used  

the same solution "frontend mem=1024M" as well.

 

rob

 

 

On Dec 4, 2007, at 4:46 AM, Andreas Yiotis wrote:

 

	I am pretty sure that the HD is OK but I decided to replace it
anyway

	and reinstall rocks using mem=1024M, which I found in the
discussion

	forum.

	I am not sure what me=1024M stands for but the cluster now seems
to  

	work

	OK.

	Thanks for your help.

	 

	A

	 

	 

	-----Original Message-----

	From: Greg Bruno [mailto:greg.bruno at gmail.com]

	Sent: Friday, November 30, 2007 7:10 PM

	To: Andreas Yiotis

	Cc: npaci-rocks-discussion at sdsc.edu

	Subject: Re: [Rocks-Discuss] Journal commit i/o error &
Input/Output

	error

	 

	On Nov 30, 2007 1:23 AM, Andreas Yiotis
<yiotis at ipta.demokritos.gr>

	wrote:

		 

		I have just installed rocks 4.3 on a pc with an ASUS
P5N-E SLI M/B

		 

		CPU E6750 at 1333FSB and 2GB RAM at 667Mhz.

		 

		 

		 

		My frontend appear to be working OK for some hours when
I get a

		 

		Message "Journal commit I/O error" and it practically
hangs.

		 

			From that point on I can issue commands at the
command prompt

		 

		But I get a message "Input/Output Error" (i.e. when I
write fdisk) or

		"Command not found" (i.e. when I write vi).

		 

		 

		 

		I suspect it has to do with the HD which is a SATA 500GB
WD or with

	the

		Nvidia 430 controller.

	 

	it sounds like a bad disk.

	 

	what is the output of:

	 

	    # dmesg

	 

	 - gb

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.sdsc.edu/pipermail/npaci-rocks-discussion/attachments/20080115/3feb8d20/attachment.html 


More information about the npaci-rocks-discussion mailing list